EMA DEMO

Europea Microfusioni Aerospaziali (EMA), part of the Rolls Royce group is a precision investment castings foundry for the production of turbine blades, vanes and component for the most modern jet-engines to civil and defence aerospace and power generation engines. EMA’s products are obtained using is the “lost wax” process and using mainly Ni-based superalloys. Ceramic inclusions problems are particularly relevant for the investment casting technology and represent one of the main sources of scraps in foundries. Therefore, the aim of the EMA’s demonstrator is to achieve more effective control on the ceramic shell mould and particularly a more effective correction of the ceramic primary slurry that ensure the best final quality of the superalloy components in the long run. The IFaCOM focus was restricted only to the shell making process, mainly targeting the evaluation of the characteristics of the ceramic primary slurry and the final status of the ceramic shell, taking in account the final status of the superalloy components. Adapting the IFaCOM approach to the EMA’s shell making process is challenging due to the fact that several measurements can be only performed offline and/or at the end of the full investment casting process; moreover, all the time scales of data acquisition are considerably different, ranging from a few minutes (for straightforward on-site automatic measurements), to a few hours (for off-site chemical laboratory surveys) and few days for final characteristics of the shell mould. All these things considered, the IFaCOM strategy in the EMA’s case involved: • a robust control of the industrial manufacturing of the ceramic shells of one aeronautical vane component in the production line, by means of a focused monitoring of the primary slurry parameters and additional control of the ceramic shell quality that are not normally actuated during standard production cycles. • in-line and an off-line data process acquisition (primary slurry parameters, shell mechanical characteristics and inclusion scrap rate of the components) and storage in a dedicated database. • statistical and cognitive systems: Statistical Process Control (SPC) & Neural Networks data analysis with the aim to find the correlations between the measured Key Process Variables and the Target Variable (output quality parameter) represented by the inclusion scrap rate. In the EMA’s case, due to the complexity and the size of the production plant and machineries, practically no test could be performed in the laboratory and activities; measurements and analyses had to be carefully planned and carried out directly on the real (working) production line, while real parts were produced for the customers. With all these constraints, the IFaCOM approach was tailored to the EMA’s case showing both long-term and short-term control loops • Short-term loop: real-time control of the shell making process through the introduction of new sensors and new equipment, the automation of manual measurement processes and the integration of all the solutions. Several sensors have been employed for the acquisition of some physical-chemical parameters on the primary equiassic slurry such as viscosity, temperature, plate weight (indirect measurement of the density). • Long-term loop: process long-term optimization system based on a neural network approach. This approach relies on the continuously updated sensor data coming from the in-line sensors mentioned above and the post process analyses (silica content, shell mechanical hardness, adhesion of primary to secondary shell layers, pH, etc.). The neural network analyser developed within the project constantly evaluates and updates the correlation between product quality and the primary slurry condition. This allows to define the reference part quality and to extrapolate the optimal slurry characteristics suggesting possible corrections to the slurry composition. As mentioned above, this process is dynamic, and allows a long-term optimization of the product quality based on a constantly updated dataset and an increasingly more robust correlation and prediction capabilities The IFaCOM system implemented in the EMA’s demonstrator performs a hybrid data collection (partly online, partly offline) and data storage is done on a dedicated SQL database, developed during the project. After the acquisition, a sensor vector (a .csv file, with all data) is generated. This file represents the input for the successive IFaCOM Processing Algorithms segment. On the grounds of a neural network analysis, this module predicts optimal values for the slurry parameters to be processed for the successive mould generation. The suggested values and the related corrections, based both on indications provided by the neural network analysis and by EMA’s tested internal formulae, are presented on a PC panel through the IFaCOM GUI. The visualization of the classical correction allows to compare standard values with neural networks’ suggestions in order to double check artificial intelligence outputs and giving the operator the possibility to choose the best correction values for the primary slurry.